ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 01:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

bring back some inline asm

Browse Source
This commit is contained in:
Chao Liu
2019-06-17 17:28:24 -05:00
parent 9d59a39af3
commit 3c0646d490
2 changed files with 73 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
composable_kernel/include/utility/amd_inline_asm.hpp
View File

@@ -5,6 +5,79 @@
namespace ck {
// cast a pointer of LDS to its address
extern "C" __attribute__((address_space(3))) void* __to_local(void* p)[[hc]];
__device__ void vmcnt(index_t cnt)
{
if(cnt == 0)
{
asm volatile("\n \
s_waitcnt vmcnt(0) \n \
" ::);
}
else if(cnt == 1)
{
asm volatile("\n \
s_waitcnt vmcnt(1) \n \
" ::);
}
else if(cnt == 2)
{
asm volatile("\n \
s_waitcnt vmcnt(2) \n \
" ::);
}
else if(cnt == 4)
{
asm volatile("\n \
s_waitcnt vmcnt(2) \n \
" ::);
}
else
{
assert(false);
}
}
__device__ void lgkmcnt(index_t cnt)
{
if(cnt == 0)
{
asm volatile("\n \
s_waitcnt lgkmcnt(0) \n \
" ::);
}
else if(cnt == 1)
{
asm volatile("\n \
s_waitcnt lgkmcnt(1) \n \
" ::);
}
else if(cnt == 2)
{
asm volatile("\n \
s_waitcnt lgkmcnt(2) \n \
" ::);
}
else if(cnt == 3)
{
asm volatile("\n \
s_waitcnt lgkmcnt(3) \n \
" ::);
}
else if(cnt == 4)
{
asm volatile("\n \
s_waitcnt lgkmcnt(4) \n \
" ::);
}
else
{
assert(false);
}
}
__device__ void outerProduct1x4(const float* a, const float* b, float* c)
{
asm volatile("\n \

15
composable_kernel/include/utility/config_amd.hpp.in
View File

@@ -24,21 +24,6 @@ __device__ void fused_multiply_accumulate(float& d, const float& s0, const float
d += s0 * s1;
}
#if 0
__device__ void fused_multiply_accumulate(half& d, const half& s0, const half& s1) { d += s0 * s1; }
__device__ void fused_multiply_accumulate(half& d, const half2& s0, const half2& s1)
{
d += s0.x * s1.x;
d += s0.y * s1.y;
}
__device__ void fused_multiply_accumulate(float& d, const half2& s0, const half2& s1)
{
d += s0.x * s1.x + s0.y * s1.y;
}
#endif
} // namespace ck
#endif